EUVIMEDEUVIMED entdecken
Crossref

Security in Multi-Agent AI Systems: Modeling Emergent Vulnerabilities via Trust Graphs

Harsh Verma

International Journal of Scientific Research and Management (IJSRM) · 2026 · 14 (04) · 2863-2873

DOI: 10.18535/ijsrm/v14i04.ec04 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Scientific Research and Management (IJSRM), 2026. Verfasst von Harsh Verma.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Autonomous multi-agent artificial intelligence (AI) systems have emerged as a rapidly evolving field that revolutionizes the way autonomous systems can make decisions together, collaborate on tasks, and learn, thereby opening new paradigms for distributed decision-making, task execution, and adaptive learning. The complexity of these systems, however, brings in several serious security issues due to the dependency of the autonomous agents. In contrast, vulnerabilities in multi-agent systems are no longer bound to their individual agents, but are spread across the system through their trust-based interactions, leading to systemic risks that can't be adequately identified or protected using traditional security methods. These emerging risks underscore the importance of a structured approach that can account for trust as a fluid and security-sensitive factor in interdependent AI systems. In order to tackle this challenge, this paper proposes a novel graph-based approach to the representation of trust relationships and vulnerability propagation in multi-agent AI systems, called the Emergent Trust Vulnerability Graph (ETVG). Furthermore, the study introduces an Agentic Trust Dynamics Theory (ATDT) as a conceptual theory to model the evolution, loss, and impact of trust on emerging security vulnerabilities in an autonomous agent network. These contributions form a framework for structuring the analysis of systemic risks that emerge from interactions between agents, and not from the failure of individual agents. Methodologically, the proposed framework utilizes the most commonly used graph modeling techniques in which an agent is modeled as a node and trust is modeled as a weight on the edges. The influence and compromise propagation through connected agents is simulated using trust propagation analysis. Controlled simulation experiments are conducted, utilizing multi-agent coordination environments with different degrees of adversarial interference and trust manipulation scenarios, in order to validate the model. The experimental results show that the ETVG framework achieves a high accuracy in vulnerability detection and mitigates the cascading effect of compromise across agent networks. In addition, the proposed model improves the capability of trust-risk prediction, which is able to discover high-risk nodes in complex interaction structures effectively. The framework exhibits better performance in maintaining the stability of the system under attack, as compared with conventional baseline approaches. This work is important for the understanding of the fundamental problems of security in autonomous AI environments. This study offers new means to design a resilient multi-agent system that is able to resist emergent threats by formalizing trust as a dynamic security vector. The results also point to promising avenues for future research in developing secure, adaptive, and self-regulating infrastructures for AI agents

Zitieren

APA

Verma, H. (2026). Security in Multi-Agent AI Systems: Modeling Emergent Vulnerabilities via Trust Graphs. International Journal of Scientific Research and Management (IJSRM), 14(04), 2863-2873. https://doi.org/10.18535/ijsrm/v14i04.ec04

Vancouver

Verma H. Security in Multi-Agent AI Systems: Modeling Emergent Vulnerabilities via Trust Graphs. International Journal of Scientific Research and Management (IJSRM). 2026;14(04):2863-2873. doi:10.18535/ijsrm/v14i04.ec04

BibTeX
@article{Verma2026,
  author = {Verma, Harsh},
  title = {Security in Multi-Agent AI Systems: Modeling Emergent Vulnerabilities via Trust Graphs},
  journal = {International Journal of Scientific Research and Management (IJSRM)},
  year = {2026},
  volume = {14},
  number = {04},
  pages = {2863--2873},
  doi = {10.18535/ijsrm/v14i04.ec04}
}
Kostenlos starten →